Many research papers try to explain factors influencing entrepreneurial activity, especially with regards to nascent (use variable SUBOANW) or early-stage entrepreneurial activity (use variable TEAYY). It makes sense to apply logistic regressions for these kinds of analyses. However, for many countries the prevalence of, for example, nascent entrepreneurship, is so low that so-called ‘rare events’ regressions may need to be applied. 

GEM data are especially fit for multilevel regressions, in which the probability of an event (involvement in a specific type/phase of entrepreneurship) is explained by individual characteristics and regional/national level characteristics. An example can be found in Autio and Acs (2010), examining the impact of Intellectual Property Rights on entrepreneurial activity.

It is essential to note that GEM’s data collection design is a cross-sectional one. This means that interpretations of findings may not be exclusively one-dimensional (as hypothesized many times). For instance, perceptions to entrepreneurship may change once an individual has become acquainted to entrepreneurship – for instance after an individual decided to proceed with starting a business.  Hence, a found positive relationship between perceived skills and the probability of being a (nascent) entrepreneur may not be purely interpreted as a causal mechanism in the sense that people with positive perceptions about their skills have higher probabilities to be engaged in entrepreneurial activity. 

A counter argument could be that, as the individual became acquainted with entrepreneurship in the pre-startup phase (and perhaps met others who were starting a business in doing so), s/he found out that the skills were sufficient. Put differently, the fact that a person was involved in nascent entrepreneurship may have affected particular perceptions to entrepreneurship. Entrepreneurial perceptions are dealt with in the next section.